[internship project] condition where BLAST > BWA.

Question

[Pierre Lindenbaum]

I've asked my student to write a program to convert BLAST/XML to BAM and to compare the results (quality, not speed) with BWA .

For now, the results are quite similar :-) ( testing with viral genomes , nanopore data,... )

Do you know any dataset (in SRA ...) or condition where Blast could outperform Bwa ?

Thanks.

P.

[lh3]

Blast is in theory more sensitive to highly divergent hits IF you use the right parameters.

Answer 1

[Michael]

Are you referring to NGS applications only?

• Maybe for PacBio reads: high error rates, and long few sequences.
• to detect contaminants or partial sequences
• I'd sometimes use only the unmapped portion of reads and blast against NT (how long does the internship last, to give enough time to blast millions of reads?)
• anything meta-genomics, especially viral metagenomics, sometimes done using blastx against viral proteins
• once we have assembled contigs

[Pierre Lindenbaum]

Are you referring to NGS applications only?

Currently, yes.

1. PacBio: Yes, nice idea.
2. contaminants: Yes but I would need some FASTQS known to be contaminated.

Answer 2

[Brian Bushnell]

Cross-species (or more distantly-related) mapping should favor BLAST. Human to mouse, for example.